2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2000-2004
5 * Poul-Henning Kamp. All rights reserved.
6 * Copyright (c) 1989, 1992-1993, 1995
7 * The Regents of the University of California. All rights reserved.
9 * This code is derived from software donated to Berkeley by
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * @(#)kernfs_vnops.c 8.15 (Berkeley) 5/21/95
34 * From: FreeBSD: src/sys/miscfs/kernfs/kernfs_vnops.c 1.43
44 #include <sys/param.h>
45 #include <sys/systm.h>
47 #include <sys/dirent.h>
48 #include <sys/eventhandler.h>
49 #include <sys/fcntl.h>
51 #include <sys/filedesc.h>
52 #include <sys/filio.h>
54 #include <sys/kernel.h>
56 #include <sys/malloc.h>
58 #include <sys/mount.h>
59 #include <sys/namei.h>
64 #include <sys/sysctl.h>
66 #include <sys/ttycom.h>
67 #include <sys/unistd.h>
68 #include <sys/vnode.h>
70 static struct vop_vector devfs_vnodeops;
71 static struct vop_vector devfs_specops;
72 static struct fileops devfs_ops_f;
74 #include <fs/devfs/devfs.h>
75 #include <fs/devfs/devfs_int.h>
77 #include <security/mac/mac_framework.h>
80 #include <vm/vm_extern.h>
81 #include <vm/vm_object.h>
83 static MALLOC_DEFINE(M_CDEVPDATA, "DEVFSP", "Metainfo for cdev-fp data");
85 struct mtx devfs_de_interlock;
86 MTX_SYSINIT(devfs_de_interlock, &devfs_de_interlock, "devfs interlock", MTX_DEF);
87 struct sx clone_drain_lock;
88 SX_SYSINIT(clone_drain_lock, &clone_drain_lock, "clone events drain lock");
89 struct mtx cdevpriv_mtx;
90 MTX_SYSINIT(cdevpriv_mtx, &cdevpriv_mtx, "cdevpriv lock", MTX_DEF);
92 SYSCTL_DECL(_vfs_devfs);
94 static int devfs_dotimes;
95 SYSCTL_INT(_vfs_devfs, OID_AUTO, dotimes, CTLFLAG_RW,
96 &devfs_dotimes, 0, "Update timestamps on DEVFS with default precision");
99 * Update devfs node timestamp. Note that updates are unlocked and
100 * stat(2) could see partially updated times.
103 devfs_timestamp(struct timespec *tsp)
111 if (tsp->tv_sec != ts) {
119 devfs_fp_check(struct file *fp, struct cdev **devp, struct cdevsw **dswp,
123 *dswp = devvn_refthread(fp->f_vnode, devp, ref);
124 if (*devp != fp->f_data) {
126 dev_relthread(*devp, *ref);
129 KASSERT((*devp)->si_refcount > 0,
130 ("devfs: un-referenced struct cdev *(%s)", devtoname(*devp)));
133 curthread->td_fpop = fp;
138 devfs_get_cdevpriv(void **datap)
141 struct cdev_privdata *p;
144 fp = curthread->td_fpop;
150 *datap = p->cdpd_data;
157 devfs_set_cdevpriv(void *priv, d_priv_dtor_t *priv_dtr)
160 struct cdev_priv *cdp;
161 struct cdev_privdata *p;
164 fp = curthread->td_fpop;
167 cdp = cdev2priv((struct cdev *)fp->f_data);
168 p = malloc(sizeof(struct cdev_privdata), M_CDEVPDATA, M_WAITOK);
170 p->cdpd_dtr = priv_dtr;
172 mtx_lock(&cdevpriv_mtx);
173 if (fp->f_cdevpriv == NULL) {
174 LIST_INSERT_HEAD(&cdp->cdp_fdpriv, p, cdpd_list);
176 mtx_unlock(&cdevpriv_mtx);
179 mtx_unlock(&cdevpriv_mtx);
180 free(p, M_CDEVPDATA);
187 devfs_destroy_cdevpriv(struct cdev_privdata *p)
190 mtx_assert(&cdevpriv_mtx, MA_OWNED);
191 KASSERT(p->cdpd_fp->f_cdevpriv == p,
192 ("devfs_destoy_cdevpriv %p != %p", p->cdpd_fp->f_cdevpriv, p));
193 p->cdpd_fp->f_cdevpriv = NULL;
194 LIST_REMOVE(p, cdpd_list);
195 mtx_unlock(&cdevpriv_mtx);
196 (p->cdpd_dtr)(p->cdpd_data);
197 free(p, M_CDEVPDATA);
201 devfs_fpdrop(struct file *fp)
203 struct cdev_privdata *p;
205 mtx_lock(&cdevpriv_mtx);
206 if ((p = fp->f_cdevpriv) == NULL) {
207 mtx_unlock(&cdevpriv_mtx);
210 devfs_destroy_cdevpriv(p);
214 devfs_clear_cdevpriv(void)
218 fp = curthread->td_fpop;
225 * On success devfs_populate_vp() returns with dmp->dm_lock held.
228 devfs_populate_vp(struct vnode *vp)
230 struct devfs_dirent *de;
231 struct devfs_mount *dmp;
234 ASSERT_VOP_LOCKED(vp, "devfs_populate_vp");
236 dmp = VFSTODEVFS(vp->v_mount);
237 locked = VOP_ISLOCKED(vp);
239 sx_xlock(&dmp->dm_lock);
242 /* Can't call devfs_populate() with the vnode lock held. */
246 sx_xunlock(&dmp->dm_lock);
247 vn_lock(vp, locked | LK_RETRY);
248 sx_xlock(&dmp->dm_lock);
249 if (DEVFS_DMP_DROP(dmp)) {
250 sx_xunlock(&dmp->dm_lock);
251 devfs_unmount_final(dmp);
254 if ((vp->v_iflag & VI_DOOMED) != 0) {
255 sx_xunlock(&dmp->dm_lock);
260 ("devfs_populate_vp: vp->v_data == NULL but vnode not doomed"));
261 if ((de->de_flags & DE_DOOMED) != 0) {
262 sx_xunlock(&dmp->dm_lock);
270 devfs_vptocnp(struct vop_vptocnp_args *ap)
272 struct vnode *vp = ap->a_vp;
273 struct vnode **dvp = ap->a_vpp;
274 struct devfs_mount *dmp;
275 char *buf = ap->a_buf;
276 int *buflen = ap->a_buflen;
277 struct devfs_dirent *dd, *de;
280 dmp = VFSTODEVFS(vp->v_mount);
282 error = devfs_populate_vp(vp);
289 if (vp->v_type == VCHR) {
290 i -= strlen(dd->de_cdp->cdp_c.si_name);
295 bcopy(dd->de_cdp->cdp_c.si_name, buf + i,
296 strlen(dd->de_cdp->cdp_c.si_name));
298 } else if (vp->v_type == VDIR) {
299 if (dd == dmp->dm_rootdir) {
304 i -= dd->de_dirent->d_namlen;
309 bcopy(dd->de_dirent->d_name, buf + i,
310 dd->de_dirent->d_namlen);
317 de = devfs_parent_dirent(de);
322 mtx_lock(&devfs_de_interlock);
326 mtx_unlock(&devfs_de_interlock);
332 mtx_unlock(&devfs_de_interlock);
336 sx_xunlock(&dmp->dm_lock);
341 * Construct the fully qualified path name relative to the mountpoint.
342 * If a NULL cnp is provided, no '/' is appended to the resulting path.
345 devfs_fqpn(char *buf, struct devfs_mount *dmp, struct devfs_dirent *dd,
346 struct componentname *cnp)
349 struct devfs_dirent *de;
351 sx_assert(&dmp->dm_lock, SA_LOCKED);
356 i -= cnp->cn_namelen;
360 bcopy(cnp->cn_nameptr, buf + i, cnp->cn_namelen);
362 while (de != dmp->dm_rootdir) {
363 if (cnp != NULL || i < SPECNAMELEN) {
369 i -= de->de_dirent->d_namlen;
372 bcopy(de->de_dirent->d_name, buf + i,
373 de->de_dirent->d_namlen);
374 de = devfs_parent_dirent(de);
382 devfs_allocv_drop_refs(int drop_dm_lock, struct devfs_mount *dmp,
383 struct devfs_dirent *de)
388 if (de->de_flags & DE_DOOMED)
390 if (DEVFS_DE_DROP(de)) {
391 KASSERT(not_found == 1, ("DEVFS de dropped but not doomed"));
392 devfs_dirent_free(de);
394 if (DEVFS_DMP_DROP(dmp)) {
395 KASSERT(not_found == 1,
396 ("DEVFS mount struct freed before dirent"));
398 sx_xunlock(&dmp->dm_lock);
399 devfs_unmount_final(dmp);
401 if (not_found == 1 || (drop_dm_lock && not_found != 2))
402 sx_unlock(&dmp->dm_lock);
407 devfs_insmntque_dtr(struct vnode *vp, void *arg)
409 struct devfs_dirent *de;
411 de = (struct devfs_dirent *)arg;
412 mtx_lock(&devfs_de_interlock);
415 mtx_unlock(&devfs_de_interlock);
421 * devfs_allocv shall be entered with dmp->dm_lock held, and it drops
425 devfs_allocv(struct devfs_dirent *de, struct mount *mp, int lockmode,
431 struct devfs_mount *dmp;
434 dmp = VFSTODEVFS(mp);
435 if (de->de_flags & DE_DOOMED) {
436 sx_xunlock(&dmp->dm_lock);
442 mtx_lock(&devfs_de_interlock);
446 mtx_unlock(&devfs_de_interlock);
447 sx_xunlock(&dmp->dm_lock);
448 vget(vp, lockmode | LK_INTERLOCK | LK_RETRY, curthread);
449 sx_xlock(&dmp->dm_lock);
450 if (devfs_allocv_drop_refs(0, dmp, de)) {
454 else if ((vp->v_iflag & VI_DOOMED) != 0) {
455 mtx_lock(&devfs_de_interlock);
456 if (de->de_vnode == vp) {
460 mtx_unlock(&devfs_de_interlock);
464 sx_xunlock(&dmp->dm_lock);
468 mtx_unlock(&devfs_de_interlock);
469 if (de->de_dirent->d_type == DT_CHR) {
470 if (!(de->de_cdp->cdp_flags & CDP_ACTIVE)) {
471 devfs_allocv_drop_refs(1, dmp, de);
474 dev = &de->de_cdp->cdp_c;
478 error = getnewvnode("devfs", mp, &devfs_vnodeops, &vp);
480 devfs_allocv_drop_refs(1, dmp, de);
481 printf("devfs_allocv: failed to allocate new vnode\n");
485 if (de->de_dirent->d_type == DT_CHR) {
490 /* XXX: v_rdev should be protect by vnode lock */
492 KASSERT(vp->v_usecount == 1,
493 ("%s %d (%d)\n", __func__, __LINE__, vp->v_usecount));
494 dev->si_usecount += vp->v_usecount;
495 /* Special casing of ttys for deadfs. Probably redundant. */
497 if (dsw != NULL && (dsw->d_flags & D_TTY) != 0)
498 vp->v_vflag |= VV_ISTTY;
501 if ((dev->si_flags & SI_ETERNAL) != 0)
502 vp->v_vflag |= VV_ETERNALDEV;
503 vp->v_op = &devfs_specops;
504 } else if (de->de_dirent->d_type == DT_DIR) {
506 } else if (de->de_dirent->d_type == DT_LNK) {
511 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_NOWITNESS);
513 mtx_lock(&devfs_de_interlock);
516 mtx_unlock(&devfs_de_interlock);
517 error = insmntque1(vp, mp, devfs_insmntque_dtr, de);
519 (void) devfs_allocv_drop_refs(1, dmp, de);
522 if (devfs_allocv_drop_refs(0, dmp, de)) {
527 mac_devfs_vnode_associate(mp, de, vp);
529 sx_xunlock(&dmp->dm_lock);
535 devfs_access(struct vop_access_args *ap)
537 struct vnode *vp = ap->a_vp;
538 struct devfs_dirent *de;
543 if (vp->v_type == VDIR)
546 error = vaccess(vp->v_type, de->de_mode, de->de_uid, de->de_gid,
547 ap->a_accmode, ap->a_cred, NULL);
552 p = ap->a_td->td_proc;
553 /* We do, however, allow access to the controlling terminal */
555 if (!(p->p_flag & P_CONTROLT)) {
559 if (p->p_session->s_ttydp == de->de_cdp)
565 _Static_assert(((FMASK | FCNTLFLAGS) & (FLASTCLOSE | FREVOKE)) == 0,
566 "devfs-only flag reuse failed");
569 devfs_close(struct vop_close_args *ap)
571 struct vnode *vp = ap->a_vp, *oldvp;
572 struct thread *td = ap->a_td;
574 struct cdev *dev = vp->v_rdev;
576 int dflags, error, ref, vp_locked;
579 * XXX: Don't call d_close() if we were called because of
580 * XXX: insmntque1() failure.
582 if (vp->v_data == NULL)
586 * Hack: a tty device that is a controlling terminal
587 * has a reference from the session structure.
588 * We cannot easily tell that a character device is
589 * a controlling terminal, unless it is the closing
590 * process' controlling terminal. In that case,
591 * if the reference count is 2 (this last descriptor
592 * plus the session), release the reference from the session.
597 if (vp == p->p_session->s_ttyvp) {
600 sx_xlock(&proctree_lock);
601 if (vp == p->p_session->s_ttyvp) {
602 SESS_LOCK(p->p_session);
604 if (count_dev(dev) == 2 &&
605 (vp->v_iflag & VI_DOOMED) == 0) {
606 p->p_session->s_ttyvp = NULL;
607 p->p_session->s_ttydp = NULL;
611 SESS_UNLOCK(p->p_session);
613 sx_xunlock(&proctree_lock);
620 * We do not want to really close the device if it
621 * is still in use unless we are trying to close it
622 * forcibly. Since every use (buffer, vnode, swap, cmap)
623 * holds a reference to the vnode, and because we mark
624 * any other vnodes that alias this device, when the
625 * sum of the reference counts on all the aliased
626 * vnodes descends to one, we are on last close.
628 dsw = dev_refthread(dev, &ref);
633 if (vp->v_iflag & VI_DOOMED) {
635 dflags |= FREVOKE | FNONBLOCK;
636 } else if (dsw->d_flags & D_TRACKCLOSE) {
637 /* Keep device updated on status. */
638 } else if (count_dev(dev) > 1) {
640 dev_relthread(dev, ref);
643 if (count_dev(dev) == 1)
644 dflags |= FLASTCLOSE;
647 vp_locked = VOP_ISLOCKED(vp);
649 KASSERT(dev->si_refcount > 0,
650 ("devfs_close() on un-referenced struct cdev *(%s)", devtoname(dev)));
651 error = dsw->d_close(dev, ap->a_fflag | dflags, S_IFCHR, td);
652 dev_relthread(dev, ref);
653 vn_lock(vp, vp_locked | LK_RETRY);
659 devfs_close_f(struct file *fp, struct thread *td)
665 * NB: td may be NULL if this descriptor is closed due to
666 * garbage collection from a closed UNIX domain socket.
668 fpop = curthread->td_fpop;
669 curthread->td_fpop = fp;
670 error = vnops.fo_close(fp, td);
671 curthread->td_fpop = fpop;
674 * The f_cdevpriv cannot be assigned non-NULL value while we
675 * are destroying the file.
677 if (fp->f_cdevpriv != NULL)
683 devfs_getattr(struct vop_getattr_args *ap)
685 struct vnode *vp = ap->a_vp;
686 struct vattr *vap = ap->a_vap;
687 struct devfs_dirent *de;
688 struct devfs_mount *dmp;
690 struct timeval boottime;
693 error = devfs_populate_vp(vp);
697 dmp = VFSTODEVFS(vp->v_mount);
698 sx_xunlock(&dmp->dm_lock);
701 KASSERT(de != NULL, ("Null dirent in devfs_getattr vp=%p", vp));
702 if (vp->v_type == VDIR) {
705 ("Null dir dirent in devfs_getattr vp=%p", vp));
707 vap->va_uid = de->de_uid;
708 vap->va_gid = de->de_gid;
709 vap->va_mode = de->de_mode;
710 if (vp->v_type == VLNK)
711 vap->va_size = strlen(de->de_symlink);
712 else if (vp->v_type == VDIR)
713 vap->va_size = vap->va_bytes = DEV_BSIZE;
716 if (vp->v_type != VDIR)
718 vap->va_blocksize = DEV_BSIZE;
719 vap->va_type = vp->v_type;
721 getboottime(&boottime);
724 if ((aa).tv_sec <= 3600) { \
725 (aa).tv_sec = boottime.tv_sec; \
726 (aa).tv_nsec = boottime.tv_usec * 1000; \
730 if (vp->v_type != VCHR) {
732 vap->va_atime = de->de_atime;
734 vap->va_mtime = de->de_mtime;
736 vap->va_ctime = de->de_ctime;
740 vap->va_atime = dev->si_atime;
742 vap->va_mtime = dev->si_mtime;
744 vap->va_ctime = dev->si_ctime;
746 vap->va_rdev = cdev2priv(dev)->cdp_inode;
751 vap->va_nlink = de->de_links;
752 vap->va_fileid = de->de_inode;
759 devfs_ioctl_f(struct file *fp, u_long com, void *data, struct ucred *cred, struct thread *td)
766 error = vnops.fo_ioctl(fp, com, data, cred, td);
772 fiodgname_buf_get_ptr(void *fgnp, u_long com)
775 struct fiodgname_arg fgn;
776 #ifdef COMPAT_FREEBSD32
777 struct fiodgname_arg32 fgn32;
784 return (fgnup->fgn.buf);
785 #ifdef COMPAT_FREEBSD32
787 return ((void *)(uintptr_t)fgnup->fgn32.buf);
790 panic("Unhandled ioctl command %ld", com);
795 devfs_ioctl(struct vop_ioctl_args *ap)
797 struct fiodgname_arg *fgn;
798 struct vnode *vpold, *vp;
810 dsw = devvn_refthread(vp, &dev, &ref);
813 KASSERT(dev->si_refcount > 0,
814 ("devfs: un-referenced struct cdev *(%s)", devtoname(dev)));
818 *(int *)ap->a_data = dsw->d_flags & D_TYPEMASK;
822 #ifdef COMPAT_FREEBSD32
831 error = copyout(p, fiodgname_buf_get_ptr(fgn, com), i);
834 error = dsw->d_ioctl(dev, com, ap->a_data, ap->a_fflag, td);
837 dev_relthread(dev, ref);
838 if (error == ENOIOCTL)
841 if (error == 0 && com == TIOCSCTTY) {
842 /* Do nothing if reassigning same control tty */
843 sx_slock(&proctree_lock);
844 if (td->td_proc->p_session->s_ttyvp == vp) {
845 sx_sunlock(&proctree_lock);
849 vpold = td->td_proc->p_session->s_ttyvp;
851 SESS_LOCK(td->td_proc->p_session);
852 td->td_proc->p_session->s_ttyvp = vp;
853 td->td_proc->p_session->s_ttydp = cdev2priv(dev);
854 SESS_UNLOCK(td->td_proc->p_session);
856 sx_sunlock(&proctree_lock);
858 /* Get rid of reference to old control tty */
867 devfs_kqfilter_f(struct file *fp, struct knote *kn)
877 error = devfs_fp_check(fp, &dev, &dsw, &ref);
880 error = dsw->d_kqfilter(dev, kn);
882 dev_relthread(dev, ref);
887 devfs_prison_check(struct devfs_dirent *de, struct thread *td)
889 struct cdev_priv *cdp;
897 dcr = cdp->cdp_c.si_cred;
901 error = prison_check(td->td_ucred, dcr);
904 /* We do, however, allow access to the controlling terminal */
907 if (!(p->p_flag & P_CONTROLT)) {
911 if (p->p_session->s_ttydp == cdp)
918 devfs_lookupx(struct vop_lookup_args *ap, int *dm_unlock)
920 struct componentname *cnp;
921 struct vnode *dvp, **vpp;
923 struct devfs_dirent *de, *dd;
924 struct devfs_dirent **dde;
925 struct devfs_mount *dmp;
928 int error, flags, nameiop, dvplocked;
929 char specname[SPECNAMELEN + 1], *pname;
934 pname = cnp->cn_nameptr;
936 flags = cnp->cn_flags;
937 nameiop = cnp->cn_nameiop;
939 dmp = VFSTODEVFS(mp);
943 if ((flags & ISLASTCN) && nameiop == RENAME)
946 if (dvp->v_type != VDIR)
949 if ((flags & ISDOTDOT) && (dvp->v_vflag & VV_ROOT))
952 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td);
956 if (cnp->cn_namelen == 1 && *pname == '.') {
957 if ((flags & ISLASTCN) && nameiop != LOOKUP)
964 if (flags & ISDOTDOT) {
965 if ((flags & ISLASTCN) && nameiop != LOOKUP)
967 de = devfs_parent_dirent(dd);
970 dvplocked = VOP_ISLOCKED(dvp);
972 error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK,
975 vn_lock(dvp, dvplocked | LK_RETRY);
980 de = devfs_find(dd, cnp->cn_nameptr, cnp->cn_namelen, 0);
981 while (de == NULL) { /* While(...) so we can use break */
983 if (nameiop == DELETE)
987 * OK, we didn't have an entry for the name we were asked for
988 * so we try to see if anybody can create it on demand.
990 pname = devfs_fqpn(specname, dmp, dd, cnp);
996 sx_xunlock(&dmp->dm_lock);
997 sx_slock(&clone_drain_lock);
998 EVENTHANDLER_INVOKE(dev_clone,
999 td->td_ucred, pname, strlen(pname), &cdev);
1000 sx_sunlock(&clone_drain_lock);
1003 sx_xlock(&dmp->dm_lock);
1004 else if (devfs_populate_vp(dvp) != 0) {
1006 sx_xlock(&dmp->dm_lock);
1007 if (DEVFS_DMP_DROP(dmp)) {
1008 sx_xunlock(&dmp->dm_lock);
1009 devfs_unmount_final(dmp);
1011 sx_xunlock(&dmp->dm_lock);
1015 if (DEVFS_DMP_DROP(dmp)) {
1017 sx_xunlock(&dmp->dm_lock);
1018 devfs_unmount_final(dmp);
1028 dde = &cdev2priv(cdev)->cdp_dirents[dmp->dm_idx];
1029 if (dde != NULL && *dde != NULL)
1036 if (de == NULL || de->de_flags & DE_WHITEOUT) {
1037 if ((nameiop == CREATE || nameiop == RENAME) &&
1038 (flags & (LOCKPARENT | WANTPARENT)) && (flags & ISLASTCN)) {
1039 cnp->cn_flags |= SAVENAME;
1040 return (EJUSTRETURN);
1045 if (devfs_prison_check(de, td))
1048 if ((cnp->cn_nameiop == DELETE) && (flags & ISLASTCN)) {
1049 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
1058 error = devfs_allocv(de, mp, cnp->cn_lkflags & LK_TYPE_MASK, vpp);
1064 devfs_lookup(struct vop_lookup_args *ap)
1067 struct devfs_mount *dmp;
1070 if (devfs_populate_vp(ap->a_dvp) != 0)
1073 dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1075 j = devfs_lookupx(ap, &dm_unlock);
1077 sx_xunlock(&dmp->dm_lock);
1082 devfs_mknod(struct vop_mknod_args *ap)
1084 struct componentname *cnp;
1085 struct vnode *dvp, **vpp;
1086 struct devfs_dirent *dd, *de;
1087 struct devfs_mount *dmp;
1091 * The only type of node we should be creating here is a
1092 * character device, for anything else return EOPNOTSUPP.
1094 if (ap->a_vap->va_type != VCHR)
1095 return (EOPNOTSUPP);
1097 dmp = VFSTODEVFS(dvp->v_mount);
1104 sx_xlock(&dmp->dm_lock);
1105 TAILQ_FOREACH(de, &dd->de_dlist, de_list) {
1106 if (cnp->cn_namelen != de->de_dirent->d_namlen)
1108 if (de->de_dirent->d_type == DT_CHR &&
1109 (de->de_cdp->cdp_flags & CDP_ACTIVE) == 0)
1111 if (bcmp(cnp->cn_nameptr, de->de_dirent->d_name,
1112 de->de_dirent->d_namlen) != 0)
1114 if (de->de_flags & DE_WHITEOUT)
1120 de->de_flags &= ~DE_WHITEOUT;
1121 error = devfs_allocv(de, dvp->v_mount, LK_EXCLUSIVE, vpp);
1124 sx_xunlock(&dmp->dm_lock);
1130 devfs_open(struct vop_open_args *ap)
1132 struct thread *td = ap->a_td;
1133 struct vnode *vp = ap->a_vp;
1134 struct cdev *dev = vp->v_rdev;
1135 struct file *fp = ap->a_fp;
1136 int error, ref, vlocked;
1141 if (vp->v_type == VBLK)
1147 /* Make this field valid before any I/O in d_open. */
1148 if (dev->si_iosize_max == 0)
1149 dev->si_iosize_max = DFLTPHYS;
1151 dsw = dev_refthread(dev, &ref);
1154 if (fp == NULL && dsw->d_fdopen != NULL) {
1155 dev_relthread(dev, ref);
1159 vlocked = VOP_ISLOCKED(vp);
1168 if (dsw->d_fdopen != NULL)
1169 error = dsw->d_fdopen(dev, ap->a_mode, td, fp);
1171 error = dsw->d_open(dev, ap->a_mode, S_IFCHR, td);
1172 /* Clean up any cdevpriv upon error. */
1174 devfs_clear_cdevpriv();
1177 vn_lock(vp, vlocked | LK_RETRY);
1178 dev_relthread(dev, ref);
1180 if (error == ERESTART)
1185 #if 0 /* /dev/console */
1186 KASSERT(fp != NULL, ("Could not vnode bypass device on NULL fp"));
1191 if (fp->f_ops == &badfileops)
1192 finit(fp, fp->f_flag, DTYPE_VNODE, dev, &devfs_ops_f);
1193 mtxp = mtx_pool_find(mtxpool_sleep, fp);
1196 * Hint to the dofilewrite() to not force the buffer draining
1197 * on the writer to the file. Most likely, the write would
1198 * not need normal buffers.
1201 fp->f_vnread_flags |= FDEVFS_VNODE;
1207 devfs_pathconf(struct vop_pathconf_args *ap)
1210 switch (ap->a_name) {
1211 case _PC_FILESIZEBITS:
1215 *ap->a_retval = NAME_MAX;
1218 *ap->a_retval = INT_MAX;
1220 case _PC_SYMLINK_MAX:
1221 *ap->a_retval = MAXPATHLEN;
1224 if (ap->a_vp->v_vflag & VV_ISTTY) {
1225 *ap->a_retval = MAX_CANON;
1230 if (ap->a_vp->v_vflag & VV_ISTTY) {
1231 *ap->a_retval = MAX_INPUT;
1236 if (ap->a_vp->v_vflag & VV_ISTTY) {
1237 *ap->a_retval = _POSIX_VDISABLE;
1241 case _PC_MAC_PRESENT:
1244 * If MAC is enabled, devfs automatically supports
1245 * trivial non-persistant label storage.
1252 case _PC_CHOWN_RESTRICTED:
1256 return (vop_stdpathconf(ap));
1263 devfs_poll_f(struct file *fp, int events, struct ucred *cred, struct thread *td)
1271 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1273 error = vnops.fo_poll(fp, events, cred, td);
1276 error = dsw->d_poll(dev, events, td);
1278 dev_relthread(dev, ref);
1283 * Print out the contents of a special device vnode.
1286 devfs_print(struct vop_print_args *ap)
1289 printf("\tdev %s\n", devtoname(ap->a_vp->v_rdev));
1294 devfs_read_f(struct file *fp, struct uio *uio, struct ucred *cred,
1295 int flags, struct thread *td)
1298 int ioflag, error, ref;
1303 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1306 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1308 error = vnops.fo_read(fp, uio, cred, flags, td);
1311 resid = uio->uio_resid;
1312 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT);
1313 if (ioflag & O_DIRECT)
1314 ioflag |= IO_DIRECT;
1316 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1317 error = dsw->d_read(dev, uio, ioflag);
1318 if (uio->uio_resid != resid || (error == 0 && resid != 0))
1319 devfs_timestamp(&dev->si_atime);
1321 dev_relthread(dev, ref);
1323 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1328 devfs_readdir(struct vop_readdir_args *ap)
1333 struct devfs_dirent *dd;
1334 struct devfs_dirent *de;
1335 struct devfs_mount *dmp;
1337 int *tmp_ncookies = NULL;
1339 if (ap->a_vp->v_type != VDIR)
1343 if (uio->uio_offset < 0)
1347 * XXX: This is a temporary hack to get around this filesystem not
1348 * supporting cookies. We store the location of the ncookies pointer
1349 * in a temporary variable before calling vfs_subr.c:vfs_read_dirent()
1350 * and set the number of cookies to 0. We then set the pointer to
1351 * NULL so that vfs_read_dirent doesn't try to call realloc() on
1352 * ap->a_cookies. Later in this function, we restore the ap->a_ncookies
1353 * pointer to its original location before returning to the caller.
1355 if (ap->a_ncookies != NULL) {
1356 tmp_ncookies = ap->a_ncookies;
1357 *ap->a_ncookies = 0;
1358 ap->a_ncookies = NULL;
1361 dmp = VFSTODEVFS(ap->a_vp->v_mount);
1362 if (devfs_populate_vp(ap->a_vp) != 0) {
1363 if (tmp_ncookies != NULL)
1364 ap->a_ncookies = tmp_ncookies;
1368 de = ap->a_vp->v_data;
1370 TAILQ_FOREACH(dd, &de->de_dlist, de_list) {
1371 KASSERT(dd->de_cdp != (void *)0xdeadc0de, ("%s %d\n", __func__, __LINE__));
1372 if (dd->de_flags & (DE_COVERED | DE_WHITEOUT))
1374 if (devfs_prison_check(dd, uio->uio_td))
1376 if (dd->de_dirent->d_type == DT_DIR)
1381 MPASS(dp->d_reclen == GENERIC_DIRSIZ(dp));
1382 if (dp->d_reclen > uio->uio_resid)
1384 dp->d_fileno = de->de_inode;
1385 /* NOTE: d_off is the offset for the *next* entry. */
1386 dp->d_off = off + dp->d_reclen;
1387 if (off >= uio->uio_offset) {
1388 error = vfs_read_dirent(ap, dp, off);
1392 off += dp->d_reclen;
1394 sx_xunlock(&dmp->dm_lock);
1395 uio->uio_offset = off;
1398 * Restore ap->a_ncookies if it wasn't originally NULL in the first
1401 if (tmp_ncookies != NULL)
1402 ap->a_ncookies = tmp_ncookies;
1408 devfs_readlink(struct vop_readlink_args *ap)
1410 struct devfs_dirent *de;
1412 de = ap->a_vp->v_data;
1413 return (uiomove(de->de_symlink, strlen(de->de_symlink), ap->a_uio));
1417 devfs_reclaim(struct vop_reclaim_args *ap)
1420 struct devfs_dirent *de;
1423 mtx_lock(&devfs_de_interlock);
1426 de->de_vnode = NULL;
1429 mtx_unlock(&devfs_de_interlock);
1430 vnode_destroy_vobject(vp);
1435 devfs_reclaim_vchr(struct vop_reclaim_args *ap)
1441 MPASS(vp->v_type == VCHR);
1450 dev->si_usecount -= vp->v_usecount;
1459 devfs_remove(struct vop_remove_args *ap)
1461 struct vnode *dvp = ap->a_dvp;
1462 struct vnode *vp = ap->a_vp;
1463 struct devfs_dirent *dd;
1464 struct devfs_dirent *de, *de_covered;
1465 struct devfs_mount *dmp = VFSTODEVFS(vp->v_mount);
1467 ASSERT_VOP_ELOCKED(dvp, "devfs_remove");
1468 ASSERT_VOP_ELOCKED(vp, "devfs_remove");
1470 sx_xlock(&dmp->dm_lock);
1471 dd = ap->a_dvp->v_data;
1473 if (de->de_cdp == NULL) {
1474 TAILQ_REMOVE(&dd->de_dlist, de, de_list);
1475 if (de->de_dirent->d_type == DT_LNK) {
1476 de_covered = devfs_find(dd, de->de_dirent->d_name,
1477 de->de_dirent->d_namlen, 0);
1478 if (de_covered != NULL)
1479 de_covered->de_flags &= ~DE_COVERED;
1481 /* We need to unlock dvp because devfs_delete() may lock it. */
1485 devfs_delete(dmp, de, 0);
1486 sx_xunlock(&dmp->dm_lock);
1488 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1489 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1491 de->de_flags |= DE_WHITEOUT;
1492 sx_xunlock(&dmp->dm_lock);
1498 * Revoke is called on a tty when a terminal session ends. The vnode
1499 * is orphaned by setting v_op to deadfs so we need to let go of it
1500 * as well so that we create a new one next time around.
1504 devfs_revoke(struct vop_revoke_args *ap)
1506 struct vnode *vp = ap->a_vp, *vp2;
1508 struct cdev_priv *cdp;
1509 struct devfs_dirent *de;
1512 KASSERT((ap->a_flags & REVOKEALL) != 0, ("devfs_revoke !REVOKEALL"));
1515 cdp = cdev2priv(dev);
1528 mtx_lock(&devfs_de_interlock);
1531 for (i = 0; i <= cdp->cdp_maxdirent; i++) {
1532 de = cdp->cdp_dirents[i];
1540 mtx_unlock(&devfs_de_interlock);
1541 if (vget(vp2, LK_EXCLUSIVE | LK_INTERLOCK,
1555 mtx_unlock(&devfs_de_interlock);
1560 if (!(cdp->cdp_flags & CDP_ACTIVE) && cdp->cdp_inuse == 0) {
1561 TAILQ_REMOVE(&cdevp_list, cdp, cdp_list);
1563 dev_rel(&cdp->cdp_c);
1567 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1572 devfs_rioctl(struct vop_ioctl_args *ap)
1575 struct devfs_mount *dmp;
1579 vn_lock(vp, LK_SHARED | LK_RETRY);
1580 if (vp->v_iflag & VI_DOOMED) {
1584 dmp = VFSTODEVFS(vp->v_mount);
1585 sx_xlock(&dmp->dm_lock);
1587 DEVFS_DMP_HOLD(dmp);
1588 devfs_populate(dmp);
1589 if (DEVFS_DMP_DROP(dmp)) {
1590 sx_xunlock(&dmp->dm_lock);
1591 devfs_unmount_final(dmp);
1594 error = devfs_rules_ioctl(dmp, ap->a_command, ap->a_data, ap->a_td);
1595 sx_xunlock(&dmp->dm_lock);
1600 devfs_rread(struct vop_read_args *ap)
1603 if (ap->a_vp->v_type != VDIR)
1605 return (VOP_READDIR(ap->a_vp, ap->a_uio, ap->a_cred, NULL, NULL, NULL));
1609 devfs_setattr(struct vop_setattr_args *ap)
1611 struct devfs_dirent *de;
1622 if ((vap->va_type != VNON) ||
1623 (vap->va_nlink != VNOVAL) ||
1624 (vap->va_fsid != VNOVAL) ||
1625 (vap->va_fileid != VNOVAL) ||
1626 (vap->va_blocksize != VNOVAL) ||
1627 (vap->va_flags != VNOVAL && vap->va_flags != 0) ||
1628 (vap->va_rdev != VNOVAL) ||
1629 ((int)vap->va_bytes != VNOVAL) ||
1630 (vap->va_gen != VNOVAL)) {
1634 error = devfs_populate_vp(vp);
1639 if (vp->v_type == VDIR)
1643 if (vap->va_uid == (uid_t)VNOVAL)
1647 if (vap->va_gid == (gid_t)VNOVAL)
1651 if (uid != de->de_uid || gid != de->de_gid) {
1652 if ((ap->a_cred->cr_uid != de->de_uid) || uid != de->de_uid ||
1653 (gid != de->de_gid && !groupmember(gid, ap->a_cred))) {
1654 error = priv_check(td, PRIV_VFS_CHOWN);
1663 if (vap->va_mode != (mode_t)VNOVAL) {
1664 if (ap->a_cred->cr_uid != de->de_uid) {
1665 error = priv_check(td, PRIV_VFS_ADMIN);
1669 de->de_mode = vap->va_mode;
1673 if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
1674 error = vn_utimes_perm(vp, vap, ap->a_cred, td);
1677 if (vap->va_atime.tv_sec != VNOVAL) {
1678 if (vp->v_type == VCHR)
1679 vp->v_rdev->si_atime = vap->va_atime;
1681 de->de_atime = vap->va_atime;
1683 if (vap->va_mtime.tv_sec != VNOVAL) {
1684 if (vp->v_type == VCHR)
1685 vp->v_rdev->si_mtime = vap->va_mtime;
1687 de->de_mtime = vap->va_mtime;
1693 if (vp->v_type == VCHR)
1694 vfs_timestamp(&vp->v_rdev->si_ctime);
1696 vfs_timestamp(&de->de_mtime);
1700 sx_xunlock(&VFSTODEVFS(vp->v_mount)->dm_lock);
1706 devfs_setlabel(struct vop_setlabel_args *ap)
1709 struct devfs_dirent *de;
1714 mac_vnode_relabel(ap->a_cred, vp, ap->a_label);
1715 mac_devfs_update(vp->v_mount, de, vp);
1722 devfs_stat_f(struct file *fp, struct stat *sb, struct ucred *cred, struct thread *td)
1725 return (vnops.fo_stat(fp, sb, cred, td));
1729 devfs_symlink(struct vop_symlink_args *ap)
1732 struct devfs_dirent *dd;
1733 struct devfs_dirent *de, *de_covered, *de_dotdot;
1734 struct devfs_mount *dmp;
1736 error = priv_check(curthread, PRIV_DEVFS_SYMLINK);
1739 dmp = VFSTODEVFS(ap->a_dvp->v_mount);
1740 if (devfs_populate_vp(ap->a_dvp) != 0)
1743 dd = ap->a_dvp->v_data;
1744 de = devfs_newdirent(ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen);
1745 de->de_flags = DE_USER;
1749 de->de_inode = alloc_unr(devfs_inos);
1751 de->de_dirent->d_type = DT_LNK;
1752 i = strlen(ap->a_target) + 1;
1753 de->de_symlink = malloc(i, M_DEVFS, M_WAITOK);
1754 bcopy(ap->a_target, de->de_symlink, i);
1756 mac_devfs_create_symlink(ap->a_cnp->cn_cred, dmp->dm_mount, dd, de);
1758 de_covered = devfs_find(dd, de->de_dirent->d_name,
1759 de->de_dirent->d_namlen, 0);
1760 if (de_covered != NULL) {
1761 if ((de_covered->de_flags & DE_USER) != 0) {
1762 devfs_delete(dmp, de, DEVFS_DEL_NORECURSE);
1763 sx_xunlock(&dmp->dm_lock);
1766 KASSERT((de_covered->de_flags & DE_COVERED) == 0,
1767 ("devfs_symlink: entry %p already covered", de_covered));
1768 de_covered->de_flags |= DE_COVERED;
1771 de_dotdot = TAILQ_FIRST(&dd->de_dlist); /* "." */
1772 de_dotdot = TAILQ_NEXT(de_dotdot, de_list); /* ".." */
1773 TAILQ_INSERT_AFTER(&dd->de_dlist, de_dotdot, de, de_list);
1774 devfs_dir_ref_de(dmp, dd);
1775 devfs_rules_apply(dmp, de);
1777 return (devfs_allocv(de, ap->a_dvp->v_mount, LK_EXCLUSIVE, ap->a_vpp));
1781 devfs_truncate_f(struct file *fp, off_t length, struct ucred *cred, struct thread *td)
1784 return (vnops.fo_truncate(fp, length, cred, td));
1788 devfs_write_f(struct file *fp, struct uio *uio, struct ucred *cred,
1789 int flags, struct thread *td)
1792 int error, ioflag, ref;
1797 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1800 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1802 error = vnops.fo_write(fp, uio, cred, flags, td);
1805 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
1806 ioflag = fp->f_flag & (O_NONBLOCK | O_DIRECT | O_FSYNC);
1807 if (ioflag & O_DIRECT)
1808 ioflag |= IO_DIRECT;
1809 foffset_lock_uio(fp, uio, flags | FOF_NOLOCK);
1811 resid = uio->uio_resid;
1813 error = dsw->d_write(dev, uio, ioflag);
1814 if (uio->uio_resid != resid || (error == 0 && resid != 0)) {
1815 devfs_timestamp(&dev->si_ctime);
1816 dev->si_mtime = dev->si_ctime;
1819 dev_relthread(dev, ref);
1821 foffset_unlock_uio(fp, uio, flags | FOF_NOLOCK | FOF_NEXTOFF);
1826 devfs_mmap_f(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1827 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1842 * Ensure that file and memory protections are
1846 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1847 maxprot = VM_PROT_NONE;
1848 if ((prot & VM_PROT_EXECUTE) != 0)
1851 maxprot = VM_PROT_EXECUTE;
1852 if ((fp->f_flag & FREAD) != 0)
1853 maxprot |= VM_PROT_READ;
1854 else if ((prot & VM_PROT_READ) != 0)
1858 * If we are sharing potential changes via MAP_SHARED and we
1859 * are trying to get write permission although we opened it
1860 * without asking for it, bail out.
1862 * Note that most character devices always share mappings.
1863 * The one exception is that D_MMAP_ANON devices
1864 * (i.e. /dev/zero) permit private writable mappings.
1866 * Rely on vm_mmap_cdev() to fail invalid MAP_PRIVATE requests
1867 * as well as updating maxprot to permit writing for
1868 * D_MMAP_ANON devices rather than doing that here.
1870 if ((flags & MAP_SHARED) != 0) {
1871 if ((fp->f_flag & FWRITE) != 0)
1872 maxprot |= VM_PROT_WRITE;
1873 else if ((prot & VM_PROT_WRITE) != 0)
1876 maxprot &= cap_maxprot;
1879 error = devfs_fp_check(fp, &dev, &dsw, &ref);
1883 error = vm_mmap_cdev(td, size, prot, &maxprot, &flags, dev, dsw, &foff,
1886 dev_relthread(dev, ref);
1890 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1893 vm_object_deallocate(object);
1898 dev2udev(struct cdev *x)
1902 return (cdev2priv(x)->cdp_inode);
1905 static struct fileops devfs_ops_f = {
1906 .fo_read = devfs_read_f,
1907 .fo_write = devfs_write_f,
1908 .fo_truncate = devfs_truncate_f,
1909 .fo_ioctl = devfs_ioctl_f,
1910 .fo_poll = devfs_poll_f,
1911 .fo_kqfilter = devfs_kqfilter_f,
1912 .fo_stat = devfs_stat_f,
1913 .fo_close = devfs_close_f,
1914 .fo_chmod = vn_chmod,
1915 .fo_chown = vn_chown,
1916 .fo_sendfile = vn_sendfile,
1918 .fo_fill_kinfo = vn_fill_kinfo,
1919 .fo_mmap = devfs_mmap_f,
1920 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
1923 /* Vops for non-CHR vnodes in /dev. */
1924 static struct vop_vector devfs_vnodeops = {
1925 .vop_default = &default_vnodeops,
1927 .vop_access = devfs_access,
1928 .vop_getattr = devfs_getattr,
1929 .vop_ioctl = devfs_rioctl,
1930 .vop_lookup = devfs_lookup,
1931 .vop_mknod = devfs_mknod,
1932 .vop_pathconf = devfs_pathconf,
1933 .vop_read = devfs_rread,
1934 .vop_readdir = devfs_readdir,
1935 .vop_readlink = devfs_readlink,
1936 .vop_reclaim = devfs_reclaim,
1937 .vop_remove = devfs_remove,
1938 .vop_revoke = devfs_revoke,
1939 .vop_setattr = devfs_setattr,
1941 .vop_setlabel = devfs_setlabel,
1943 .vop_symlink = devfs_symlink,
1944 .vop_vptocnp = devfs_vptocnp,
1947 /* Vops for VCHR vnodes in /dev. */
1948 static struct vop_vector devfs_specops = {
1949 .vop_default = &default_vnodeops,
1951 .vop_access = devfs_access,
1952 .vop_bmap = VOP_PANIC,
1953 .vop_close = devfs_close,
1954 .vop_create = VOP_PANIC,
1955 .vop_fsync = vop_stdfsync,
1956 .vop_getattr = devfs_getattr,
1957 .vop_ioctl = devfs_ioctl,
1958 .vop_link = VOP_PANIC,
1959 .vop_mkdir = VOP_PANIC,
1960 .vop_mknod = VOP_PANIC,
1961 .vop_open = devfs_open,
1962 .vop_pathconf = devfs_pathconf,
1963 .vop_poll = dead_poll,
1964 .vop_print = devfs_print,
1965 .vop_read = dead_read,
1966 .vop_readdir = VOP_PANIC,
1967 .vop_readlink = VOP_PANIC,
1968 .vop_reallocblks = VOP_PANIC,
1969 .vop_reclaim = devfs_reclaim_vchr,
1970 .vop_remove = devfs_remove,
1971 .vop_rename = VOP_PANIC,
1972 .vop_revoke = devfs_revoke,
1973 .vop_rmdir = VOP_PANIC,
1974 .vop_setattr = devfs_setattr,
1976 .vop_setlabel = devfs_setlabel,
1978 .vop_strategy = VOP_PANIC,
1979 .vop_symlink = VOP_PANIC,
1980 .vop_vptocnp = devfs_vptocnp,
1981 .vop_write = dead_write,
1985 * Our calling convention to the device drivers used to be that we passed
1986 * vnode.h IO_* flags to read()/write(), but we're moving to fcntl.h O_
1987 * flags instead since that's what open(), close() and ioctl() takes and
1988 * we don't really want vnode.h in device drivers.
1989 * We solved the source compatibility by redefining some vnode flags to
1990 * be the same as the fcntl ones and by sending down the bitwise OR of
1991 * the respective fcntl/vnode flags. These CTASSERTS make sure nobody
1992 * pulls the rug out under this.
1994 CTASSERT(O_NONBLOCK == IO_NDELAY);
1995 CTASSERT(O_FSYNC == IO_SYNC);